** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **

[Mauller]

37mm^2 is over twice the size of the biggest GPU Nvidia has ever made so... I'm going with nope

Yes, they need different pinout, but also a different controller. Midrange will get made with a gddr5 controller, high end with hbm. I'm guessing there, might be low end gddr5, midrange/high end hbm.

If we are talking about a very large chip that is bigger than anything Nvidia has made, then there is a possibility that it is a full interposer package.

The interposer package would have fewer pins than the bare GPU since only the gpu core interfaces directly with the memory. so having 2000 odd pins sounds reasonable for external communication and power.

 

[bru]

37mm^2 is over twice the size of the biggest GPU Nvidia has ever made so... I'm going with nope

Ok now I'm confused and it would seem so are you.

This article. seems to be saying that 37mm^2 is a small die size and they expect the bigger core to be 55mm^2.

Help us out here DM, because even though we have argued in the past, I still value your knowledge in this field.

The shipment information suggests that Nvidia's GP104 Pascal GPU will be 37.5 x 37.5 mm and will feature 2,152 pins. While there's no firm information to confirm such a move, the surprisingly small die size indicates Nvidia will not use high-bandwidth memory (HBM) along with this GPU.

The reason why it appears Nvidia will not use HBM with the GP104 is that a die of this size could not accommodate both a high-performance GPU and the HBM chips, which sit on the same substrate. Instead, current speculation suggests GP104 graphics cards will use GDDR5 or GDDR5X, resulting in less memory bandwidth than a HBM solution.

Meanwhile, the top-end GP100 GPU is expected to come in at 55 x 55mm, which is larger than the Maxwell-based GM200 (seen above). Factoring in a manufacturing process shrink to 16nm, GP100's extra die size will likely be used to accommodate HBM, making it Nvidia's first GPU to use the technology.

 

[drunkenmaster]

Sorry, powers is a pain in normal text, it was really (37mm)^2 in reality. A die that has sides of 37mm length would be over 1300mm^2, Titan is what, around 550mm^2.

What they are saying is the package size of 37mm sides is the green area the die is attached to, if you look at the package for say a Maxwell GM104, you'll see the die is only around 1/4 of the size of the package it sits on. So a 37mm sided package with a 1300mm^2 size indicates a chip of around 300-350mm^2, or the midrange Pascal.

The 55mm sides indicates a package size of over 3000mm^2, this doesn't indicate a chip of around 750mm^2, it indicates a chip that is on an interposer. Basically package size of 37mm sides makes sense for midrange Pascal, a package size of 40-45mm^2 would make sense for a die like big Maxwell, but 55mm is so big it means it's got to be using an interposer as well.

 

[D3K]

I recognise some of those words.

 

[D.P.]

Ok now I'm confused and it would seem so are you.

This article. seems to be saying that 37mm^2 is a small die size and they expect the bigger core to be 55mm^2.

Help us out here DM, because even though we have argued in the past, I still value your knowledge in this field.

The actually chip to get bounded to the PCB (or interposer in this case) has a side length of 37mm, but the silicon die is smaller.

Here is an example:http://cdn.wccftech.com/wp-content/uploads/2013/03/nvidia-tegra.jpg
The small inner silver square of the actual silicon, the package is the green thing. The package size is much bigger than the core.

 

[Simon Belmont]

I remember Nvidia was quiet last time before maxwell release, but wow they haven't said one word on what's going on with Pascal at all? It's so quiet it's spooky

I hope when AMD release there new series, Nvidia will then go BAM!! PASCAL! lol

Something has got to give?

 

[bru]

I wouldn't be at all surprised if at GTC they end up on stage showing off a new card and saying it will be available in X days/weeks.

 

[Kaapstad]

I wouldn't be at all surprised if at GTC they end up on stage showing off a new card and saying .

Fixed.

 

[madmonkey]

I remember Nvidia was quiet last time before maxwell release, but wow they haven't said one word on what's going on with Pascal at all? It's so quiet it's spooky

I hope when AMD release there new series, Nvidia will then go BAM!! PASCAL! lol

Nvidia don't really need to say much at the moment, speculation is already rife and anticipation high with Pascal, without them revealing too much.

With there poor market share, AMD don't really have much choice but to do all the early running with announcements and such.

 

[Orangey]

Pascal will not be terribly exciting. Volta will be the real big leap.

 

[JediFragger]

Pascal will not be terribly exciting. Volta will be the real big leap.

Agreed, from the bits I've read about Volta it looks the business!! I'm going to *try* and make my Ti last until then. Shouldn't be too bad with Gsync kicking it along...

 

[TheF34RChannel]

Pascal will not be terribly exciting. Volta will be the real big leap.

I'm excited for both. Although I don't know too much about Volta, why do you say that will be the real big leap?

 

[Boomstick777]

Pascal will not be terribly exciting. Volta will be the real big leap.

Die shrink, new architecture, HBM 2.0. If that doesn't peak some interest / excitement for peeps who like GPU's than what will?

Full fat GP200 is gonna be pretty exiting imho, just hope we don't have to wait until Q4 2016 / Q1 2017 for it..

 

[TheF34RChannel]

Die shrink, new architecture, HBM 2.0. If that doesn't peak some interest / excitement for peeps who like GPU's than what will?

Full fat GP200 is gonna be pretty exiting imho, just hope we don't have to wait until Q4 2016 / Q1 2017 for it..

Exactly!

We probably have to... And I need something in between... The x80 most likely until the Ti comes along.

 

[Simon Belmont]

Die shrink, new architecture, HBM 2.0. If that doesn't peak some interest / excitement for peeps who like GPU's than what will?

Full fat GP200 is gonna be pretty exiting imho, just hope we don't have to wait until Q4 2016 / Q1 2017 for it..

^^This in a nut shell

 

[fenchu]

was waiting for pascal and hbm , but if its not part of this arch i will await volta and eek out titan all being well

 

[TheF34RChannel]

Out of curiosity, who will buy into Pascal before GP200 is released (and upgrade when it does, of course)?

 

[Rroff]

Could be that there isn't a (real) full fat Pascal - doesn't look like TSMC has anything above mid-power (3xxmm2) road mapped for 16nm - they seem to be shifting their focus to 10nm and beyond which could coincide with Volta depending on nVidia's timetable.

 

[TheF34RChannel]

Hmm food for thought that... Perhaps the x80 is all we'll get until Volta, and maybe that's all we need.

What do you mean by mid-power?

 

[D.P.]

Out of curiosity, who will buy into Pascal before GP200 is released (and upgrade when it does, of course)?

GP200 likely won't ever exist, we will get GP100 and then will move to Volta with GV100.